Prediction of recurrence free survival of head and neck cancer using PET/CT radiomics and clinical information

TL;DR

This study targets recurrence-free survival (RFS) prediction in oropharyngeal head and neck cancer by integrating clinical variables with multimodal radiomics derived from FDG-PET/CT within a Cox proportional hazards framework. Features are extracted from tumor regions, reduced via LASSO, and evaluated using the concordance index, with the best performance reaching a C-index of $0.74$ when combining CT, PET, and clinical data. A novel aspect is the explicit assessment of segmentation accuracy through under- and over-segmentation, revealing modality-specific sensitivities: CT features are robust to under-segmentation but degrade with over-segmentation, while PET features are more impacted by under-segmentation but can benefit from boundary information when over-segmented. Overall, multimodal radiomics improves RFS prognostication and highlights the need for careful segmentation strategies in multi-modal radiomics workflows to maximize clinical utility.

Abstract

The 5-year survival rate of Head and Neck Cancer (HNC) has not improved over the past decade and one common cause of treatment failure is recurrence. In this paper, we built Cox proportional hazard (CoxPH) models that predict the recurrence free survival (RFS) of oropharyngeal HNC patients. Our models utilise both clinical information and multimodal radiomics features extracted from tumour regions in Computed Tomography (CT) and Positron Emission Tomography (PET). Furthermore, we were one of the first studies to explore the impact of segmentation accuracy on the predictive power of the extracted radiomics features, through under- and over-segmentation study. Our models were trained using the HEad and neCK TumOR (HECKTOR) challenge data, and the best performing model achieved a concordance index (C-index) of 0.74 for the model utilising clinical information and multimodal CT and PET radiomics features, which compares favourably with the model that only used clinical information (C-index of 0.67). Our under- and over-segmentation study confirms that segmentation accuracy affects radiomics extraction, however, it affects PET and CT differently.

Figures (2)

• Figure 1: An exemplar (a) PET (b) CT and (c) Segmentation of a patient from HECKTOR challenge HECKTOR2022
• Figure 2: Radiomics workflow (Image adapted from radiomicsimg and method adapted from bourigault2021multimodal)